Chanjing-Avatar-V2V-1.3B

Chanjing-Avatar V2V 1.3B (InfiniteTalk)

Chanjing-Avatar V2V 1.3B is an audio-driven video-to-video avatar model based on Wan2.1-T2V-1.3B and the InfiniteTalk approach. It preserves source body, background, and camera motion while regenerating a face synchronized to a driving audio track.

This model card is prepared for the InfiniteTalk checkpoint release. Training, preprocessing, and inference code is available at chanjing-ai/Chanjing-Avatar-V2V-1.3B.

Chanjing-Avatar Model Family

• Chanjing-Avatar 14B: 720p image-to-video generation from a reference image and driving audio.
• Chanjing-Avatar V2V 5B: video-to-video generation that preserves source motion and regenerates the speaking face.
• Chanjing-Avatar V2V 1.3B: a lighter video-to-video model for audio-driven face animation.

The checkpoint directory is expected to contain:

Chanjing-Avatar-V2V-1.3B/
|-- model-00001-of-00002.safetensors
|-- model-00002-of-00002.safetensors
|-- model.safetensors.index.json
`-- training_init/audio_proj.safetensors

Wan2.1-T2V-1.3B, TencentGameMate/chinese-wav2vec2-base, and the face preprocessing models are required separately. Review all upstream licenses and terms before use.

Users are responsible for obtaining consent for source videos and voices and for clearly disclosing synthetic media. Do not use this model for impersonation, fraud, harassment, or deceptive content.